44 results about "Grid clustering" patented technology

The invention provides a grid clustering method and device, equipment and a storage medium, and the method comprises the steps: carrying out the layer-by-layer grid division of a target region represented by a target image according to each resolution corresponding to the target image, and obtaining N target grids corresponding to the target region; counting the grid density corresponding to each target grid to obtain N grid densities; sorting the N target grids according to the sequence of the corresponding grid densities from high to low to form a grid sequence; and clustering the N target grids according to the sequence of the N target grids in the grid sequence and the grid density corresponding to each target grid to obtain M clusters. In the embodiment of the invention, the target area is subjected to grid division through the hexagonal grid, so that the deviation and error of the clustering result can be reduced; in this way, the curvature and the non-rectangular terrain of the target area are adapted through a hexagonal grid clustering method, and the accuracy of a clustering result is improved.

The invention relates to the technical field of power distribution management, and discloses a novel power distribution network aggregation resource power prediction method and system, and the system comprises a meteorological data unit, a mutual information analysis unit, a load data unit, a load analysis unit, and a neural unit. According to the novel power prediction method and system for the aggregated resources of the power distribution network, various types of weather forecast data are obtained through a weather data unit, mutual information between actually measured wind power and weather factors and mutual information between different weather factors are calculated through a mutual information analysis unit, and then feature sorting is carried out through an incremental search method; a load data unit obtains load data and communication people flow data, space-time grid clustering and space-time coupling are carried out, user load classification, load grading and communication people flow and load relevance are analyzed, and a neural unit learns and predicts resident user loads and resident area distribution transformer loads; a load increase and decrease target is formulated, power prediction is more accurate, and the utilization rate of aggregation resources is high.

The invention discloses an unsupervised unmanned aerial vehicle air combat situation assessment method based on a double attention mechanism and a BiGRU, and belongs to the technical field of crossing of machine learning and modern military. The method comprises the following steps: firstly, designing a clustering method, providing a parameter adaptive density grid clustering method based on a fuzzy theory to process a large-scale target, and establishing an air combat information sample set with time and tactical meanings to facilitate subsequent rapid evaluation; on the basis, an encoder-decoder structure taking BiGRU as a trunk is designed to extract depth information of a historical state and a subsequent flight state of a target in two directions, and a double attention mechanism suitable for time sequence data is improved, so that the model can better pay attention to key parts and key features of an input sequence. According to the method, the generalization ability and the practical application value of an air combat situation evaluation model can be improved, the processing ability and the evaluation precision of situation data are improved, the influence of irrelevant parts in high-dimensional and long-sequence time data on an evaluation result is reduced, and the modern air combat requirement is met.

The invention discloses a self-adaptive heuristic grid clustering method suitable for non-intrusive load monitoring, which comprises the following steps of: S1, selecting low-frequency data, preprocessing the low-frequency data, and converting the selected data into a four-dimensional tensor containing total load, hour, week and month; s2, event detection is performed on the total load measurement sequence by using a time window, a four-dimensional matrix containing event amplitude and time reference is generated, events are grouped by using a heuristic grid clustering algorithm, and different clustering results represent different states of equipment in an operation process; s3, constructing a state transition matrix according to the event grouping result and the state change of each device in the operation process; and according to the energy consumption period and the state transition matrix of the equipment, reconstructing an energy consumption sequence of the equipment, completing the estimation of the energy consumption of the equipment, and calculating the use frequency of the equipment. The household data processing method has high precision and efficiency in household data processing, and can effectively separate electric energy loads of various types of electric appliances.

The invention discloses an electrodeionization system intelligent regulation and control method for a hydrogen conductivity meter, which is used for evaluating and intelligently regulating and controlling the structure state and the working state of an electrodeionization system, and comprises the following steps: carrying out automatic working mode parameter identification on the electrodeionization system based on a random subspace identification algorithm and a grid clustering algorithm; the structural state of the electrodeionization system is evaluated and intelligently regulated and controlled based on the automatic working mode parameter recognition result; based on the fact that the structural state of the electrodeionization system is evaluated as healthy or the structural state of the electrodeionization system is evaluated as healthy after intelligent regulation and control, the working state of the electrodeionization system is intelligently regulated and controlled; the working state affects the cation removal efficiency of the electrodeionization system and is related to the resin capacity of a concentrated water chamber, the resin capacity of a fresh water chamber, the ratio of the sectional area to the height of the concentrated water chamber, the ratio of the sectional area to the height of the fresh water chamber, membrane stack voltage, membrane stack current and module inlet water flow. The invention further discloses a system, electronic equipment and a computer readable storage medium.

The invention provides a frequency decoupling multi-weather scene crack identification method and system, and the method comprises the steps: carrying out the denoising of a to-be-identified concrete engineering structure picture, and obtaining a denoised image; after down-sampling is carried out on the de-noised image, a first feature is obtained through Sharing STEM processing; inputting the first feature into a plurality of frequency decoupling modules for processing, and then classifying the first feature to identify whether the concrete engineering structure picture has a crack or not; the first frequency decoupling module carries out grid division on the first feature, then carries out grid clustering and carries out position coding; high-frequency features and low-frequency features of the clustering grids with the position coding information are extracted respectively, and fusion is carried out according to relevance of the low-frequency features and the high-frequency features; and fusing the fused feature and the first feature, and inputting the fused feature and the first feature as a new first feature to a subsequent frequency decoupling module. And high-precision crack segmentation is realized under complex weather conditions, and accurate guarantee is provided for infrastructure maintenance and public traffic safety.

The invention belongs to the technical field of equipment planning, and discloses an automatic planning method for intelligent agricultural Internet of Things equipment, which comprises the following steps: dividing a target region into grid units, collecting soil moisture parameters and nutrient parameters of each grid, and establishing a regional digital twinborn model; obtaining a target crop of each grid, obtaining growth information from the crop knowledge base, and executing time sequence simulation to generate a water and fertilizer demand table of each grid; executing a water-fertilizer balance strategy based on the water-fertilizer demand table, and generating a water-fertilizer supply plan of each grid; analyzing spatial distribution characteristics of supply plans, and clustering adjacent grids with similar plans into irrigation management partitions; and generating a pipe network topological structure and an equipment planning scheme based on the partition layout. According to the method, a traditional uniformly distributed area is scientifically divided into management subareas with consistent internal requirements, so that unified management of the same subarea is realized, and the problem of supply and demand mismatching caused by neglect of spatial heterogeneity in a traditional method is solved.

The invention relates to the field of atmospheric dust pollution prevention and environmental protection, and discloses a wharf dust emission monitoring grid clustering point distribution method based on uncertainty quantization, which comprises the following steps of: a, performing dust diffusion simulation by utilizing a pollutant diffusion model, and determining a dust comprehensive evaluation concentration; b, performing comprehensive evaluation and preliminary screening based on fuzzy evaluation to obtain candidate point distribution grids; c, the candidate point distribution grids are clustered, and the candidate point distribution grids are merged into categories; and step d, determining a final dust emission monitoring point according to the category. According to the method, the objective dust comprehensive evaluation concentration is calculated by introducing the GAUSSION diffusion model based on various meteorological scenes and occurrence probabilities, and subjective judgment depending on historical experience is replaced; and a fuzzy evaluation technology and an entropy weight method are utilized to carry out mathematical quantification on various uncertain factors, so that interference of human factors can be eliminated fundamentally, and a point distribution result is more reliable and representative.

The invention discloses a trailer boundary dimension detection method based on unmanned container truck perception, and belongs to the technical field of automatic driving and vehicle engineering. In order to solve the problems of low manual correction efficiency and poor flexibility of external measurement equipment in the prior art, the invention provides the following scheme: acquiring three-dimensional point cloud information containing a vehicle through a vehicle-mounted laser radar, calculating a trailer rotation angle in real time by combining a high reflection intensity identifier, dynamically delimiting an interest region by using vehicle-mounted parameters, and filtering out a tractor point cloud; the remaining point clouds are distributed to a two-dimensional grid map, and candidate trailer point clouds are extracted through ground point cloud filtering and grid clustering analysis; finally, the maximum left width, the maximum right width, the maximum length and the maximum height of the trailer are calculated based on segmented time sequence checking, and the accurate boundary size is output. According to the method, manual intervention or external equipment support is not needed, real-time detection can be achieved under any trailer posture, and the safety and efficiency of unmanned container truck operation are effectively improved.

The invention discloses a sea route network construction method based on multistage grid clustering, and the method comprises the steps: setting a navigation region, collecting navigation data, carrying out the noise reduction, preprocessing and segmenting, and obtaining processed data; a space grid structure and a KD-tree structure are introduced on the basis of the DBSCAN algorithm to form a GBSC algorithm; a dynamic threshold mechanism is introduced on the basis of a Douglas-Peucker algorithm, and an adaptive DP algorithm is obtained; screening, clustering and correcting the trajectory segments and the trajectory points by using a GBSC algorithm and an adaptive DP algorithm to obtain a feature point set; a feature point closest to the track point is found in the feature point set through a KD-tree structure, a feature point sequence is obtained, repeated nodes are compressed to obtain a node sequence, adjacent nodes are connected, and an edge set is obtained to construct an adjacent matrix and an edge weight matrix to form an air route network; the method can truly reflect the actual navigation structure of the ship in the sea area, has a high coincidence degree with the original AIS trajectory, and significantly improves the accuracy and reliability of air route extraction.

The application discloses a weak coverage area grid clustering method and device, and the method comprises the following steps: performing slicing processing on a raster map to obtain a plurality of map slices; calculating the weak coverage probability of each map slice; screening the map slices to be clustered from the plurality of map slices according to the weak coverage probabilities of the plurality of map slices, performing clustering processing on the grids in the map slices to be clustered to obtain a weak coverage clustering cluster; collecting the update coverage data corresponding to the plurality of map slices, and updating the raster map according to the update coverage data and the weak coverage clustering cluster. The application realizes the management and processing of different slices by screening the map slices to be clustered, and does not need to perform complete traversal clustering and updating on the entire raster map, thereby saving the computing resource, effectively improving the update speed of the raster map, and solving the problems that the current clustering operation requires a large computing power and the real-time update map operation consumes a large amount of resources.

The present invention provides a method, device, and equipment for processing unmanned vehicle navigation radar echoes, belonging to the field of computer information processing technology. It solves the problems of low efficiency, unstable target echo signals, and susceptibility to sea clutter in current navigation radar echo processing. The method includes: obtaining a set of unmanned vehicle navigation radar echoes at multiple times; clustering the set of unmanned vehicle navigation radar echoes at multiple times to obtain a set of line segments; determining echo clustering data at multiple times based on the set of line segments; mapping the echo clustering data at multiple times to a grid map to obtain grid clustering data at multiple times; and clustering and merging the grid clustering data at multiple times to obtain radar echo processing data. This solution forms grid clusters by accumulating historical information, effectively improving clustering efficiency, reducing the impact of interference factors, and improving the robustness of track processing.

The invention discloses an image processing method. The method comprises the following steps: receiving a plurality of pieces of grid data of a component; transmitting the plurality of grid data to a GPU; clustering the plurality of pieces of grid data to obtain a plurality of pieces of grid clustering data; transmitting the plurality of grid clustering data to a GPU (Graphics Processing Unit); and the GPU is used for rendering the plurality of pieces of grid data and rendering the plurality of pieces of grid clustering data after receiving the plurality of pieces of grid clustering data. In the embodiment of the invention, since the grid data needs a certain time to be processed into the grid clustering data, the CPU can directly transmit a plurality of pieces of grid data to the GPU for rendering after receiving the grid data of one component, the GPU is prevented from being in an idle state in the process of generating the grid clustering data by the CPU, and the processing efficiency of the CPU is improved. Therefore, the screen cannot timely display the rendered image. After the GPU receives the grid clustering data, the grid data can be deleted, and then the grid clustering data is rendered, so that a patch after one component is dispersed can be added to a GPU video memory at one time.

The present invention discloses an obstacle recognition method based on an on-vehicle lidar, which relates to the technical field of point cloud data processing and includes: acquiring original point cloud data; performing preprocessing operations on the original point cloud data to obtain preprocessed point cloud data; performing ground segmentation on the preprocessed point cloud data to obtain obstacle point cloud data; constructing a polar coordinate grid map according to the retained obstacle point cloud data; wherein, the obstacle polar coordinate grid map includes a plurality of grids; obtaining an adaptive radial search step length for grid clustering according to a relative distance threshold; using the region growing method to label the grids that meet the requirements of the adaptive radial search step length to obtain different categories of grid sets; converting the different categories of grid sets into different categories of point cloud data sets to obtain obstacle information. The present invention can improve the recognition accuracy on the basis of ensuring the rapid recognition of obstacle targets.

The present invention discloses a point cloud clustering method applicable to driverless driving and a control method for assisting driverless driving. The clustering method includes: constructing a grid coordinate system in the horizontal plane based on the vehicle position; detecting the point cloud data packets around the vehicle, determining the grids corresponding to each point cloud, and storing its index information in the index list of the grid; binarizing the grids with an empty / non-empty index list to 0 / 1; calculating the dilation amplitude according to the preset minimum distance between classes; according to the dilation amplitude, extending outward with the grids binarized to 1 as the center, and binarizing the extended grids to 1; clustering the adjacent grids binarized to 1 together to form a convex hull; traversing the index list of each grid in each convex hull to obtain the index information of the point cloud corresponding to the same convex hull, so as to obtain the clustered point cloud cluster corresponding to the convex hull. The present invention can achieve fast clustering based on the original point cloud, realize a high-concurrency and real-time driverless algorithm, and achieve safer driving performance.

The present invention relates to a bus route planning and visual analysis method for influencing factors based on GPS data. First, city taxi order data and city point of interest data are preprocessed to obtain passenger travel origin-destination data and POI category data. Passenger boarding and alighting hotspot grids are extracted, and candidate bus stops are mined based on grid clustering. Furthermore, a multi-objective model for bus route planning is constructed, and an optimal bus route set is generated through a multi-objective optimization method. Attribute weight quantification and ranking results are obtained based on statistical route attribute information. Finally, a visual analysis system supports planners in exploring and analyzing routes in the optimal route set and supports changing route attribute weights. The system retrains based on the adjusted attribute weights or subjective ranking to update the ranking results. Through interactive exploratory analysis, an optimal bus route planning solution is obtained, providing a scientific decision-making basis and practical means for transportation departments to carry out bus route planning, thereby facilitating the development of smart cities.

The present application relates to the technical field of wind power prediction, and discloses a canyon wind power short-term power prediction method based on grid clustering, comprising the following steps: step S1, according to the characteristics of canyon wind power, the short-term power prediction method is divided into a data cleaning link, a grid decomposition link and a power prediction link; step S2, in the data cleaning link, the historical meteorological data, wind power data and wind power prediction data learned by the LSTM network are subjected to anomaly detection, data correction and smoothing processing, and complete historical data are obtained; step S3, in the grid decomposition link, the regional grid clustering of the canyon wind power station is carried out according to the latitude, and the wind power and the characteristic grid associated with the meteorological data are obtained; step S4, in the power prediction link, the power of each grid is predicted by using the LSTM network, and power superposition and error correction are carried out. The present application is used to solve the problem of low accuracy of short-term wind power prediction caused by the influence of canyon wake on canyon wind power.

The invention provides an obstacle detection method and device, electronic equipment and a vehicle. The obstacle detection method comprises the steps of obtaining obstacle information around a vehicle and an occupied grid clustering object corresponding to the obstacle information; judging whether the object occupying the grid cluster is a large vehicle or not; if the occupied grid clustering object is a large vehicle and is in a turning scene, characteristic parameters of the occupied grid clustering object are acquired, and the characteristic parameters comprise a bending point of the occupied grid clustering object and a longest axis of the occupied grid clustering object; splitting the occupied grid clustering object according to the characteristic parameters to obtain a first part and a second part; and carrying out fusion processing on the first part and the second part to obtain a tracking obstacle, thereby realizing accurate detection on an unknown type of obstacle.

The invention provides a time domain resampling grid clustering multi-task implementation method and system, and the method comprises the steps: carrying out the matching according to obtained camera data and imaging sampling moment data added to position and attitude data, calculating the sub-satellite point position of a sampling moment according to the adaptive position and attitude data, and obtaining M rows and N columns of grid angular point data; caching M rows of grid angular point data, and calculating a normal vector of a projection plane; according to the adapted position and attitude data and all camera data, calculating position and attitude data of camera data i pixel scatter points; calculating the coordinates of the pixels in the earth-fixed coordinate system; and constructing an auxiliary plane for clustering according to the M-row N-column grid angular point data and the coordinates of the normal vector and the pixel of the projection plane in the earth-fixed coordinate system. According to the invention, a grid clustering algorithm is realized in a lightweight manner, a data-driven multi-task software architecture is adopted to be deeply coupled with multi-core processor hardware, the method can be conveniently deployed in a line-on-satellite embedded system, and the method has good real-time performance.

The application discloses a regional carbon emission data monitoring and analyzing system and method, and belongs to the technical field of big data analysis and traffic carbon emission monitoring. The method comprises the following steps: acquiring multi-source data of a city road network, wherein the multi-source data at least comprises traffic gate flow data, vehicle GPS track data, weather data and signal lamp timing data; dividing a city area into a plurality of spatial grid units; based on historical vehicle flow form indexes of each grid unit, combining preset business rules reflecting city functional area characteristics, performing dynamic clustering processing on the plurality of grid units to obtain a plurality of carbon emission characteristic grid clusters; wherein the judgment basis of the business rules at least partially originates from land use property data around the grid units; for each carbon emission characteristic grid cluster, according to the correlation characteristics of its vehicle flow form characteristics and event data, selecting a corresponding prediction algorithm to establish a traffic carbon emission prediction sub-model; based on the output of each sub-model, generating a carbon emission hotspot distribution map of the city road network in a future preset time period, and associating and outputting carbon emission contribution degrees and emission attribution analysis results of each characteristic grid cluster. Through the dynamic grid clustering mechanism of "vehicle flow form clustering + city functional area rule verification", the application realizes fine monitoring and accurate tracing of city traffic mobile source emissions, and effectively improves the spatio-temporal resolution and management operability of traffic carbon emission prediction.

The present invention provides a frequency-decoupled crack identification method and system for multiple weather scenarios. The system denoises a concrete structure image to be identified to obtain a denoised image. The denoised image is downsampled and processed using SharedSTEM to obtain a first feature. The first feature is input into several frequency decoupling modules for processing and classification to identify whether cracks exist in the concrete structure image. The first frequency decoupling module divides the first feature into grids, clusters them, and performs position encoding. High-frequency and low-frequency features of the clustered grids with position encoding information are extracted and fused based on their correlation. The fused feature is then fused with the first feature and input into subsequent frequency decoupling modules as the new first feature. This ensures high-precision crack segmentation under complex weather conditions, providing precise protection for infrastructure maintenance and public transportation safety.

The invention relates to the technical field of industrial production safety monitoring, in particular to a rolling bearing vibration signal prediction method based on an improved turnip optimization network model, and the method comprises the steps: obtaining a vibration signal of a target rolling bearing, and constructing the vibration signal into a time sequence; performing variational mode decomposition on the time sequence of the vibration signal to obtain a plurality of components; the component is input into a pre-constructed signal prediction model, a prediction result of the vibration signal is output, and the signal prediction model is constructed by optimizing a bidirectional gating circulation unit based on an improved kirschner optimization algorithm. The improved squirrel optimization algorithm is constructed by introducing a Bloch unit sphere coordinate quantum bit coding mechanism, a local grid clustering selection strategy, a predator flying-away strategy and an improved multi-difference Cauchy mutation strategy into the squirrel optimization algorithm. According to the method, the generalization ability of the model is enhanced through multi-strategy optimization, and the prediction precision and robustness of the vibration signal of the rolling bearing are remarkably improved.

The present invention provides a network layout method, device, and electronic device based on spatial cluster analysis, relating to the field of network layout. The method comprises: determining the types of all target products based on historical sales data, gridding target hotspot areas, determining each subgrid area, calculating the variance value between the grid cluster center and the standard grid cluster center, processing the subgrid area using kernel density analysis to obtain clustered areas; adjusting the sales volume of each target product using the variance value and the clustered area to obtain adjusted sales volume for each target product; determining the order quantity of the target product based on the target POI; determining the fishing net area with the largest order quantity as the target network point; and determining the product configuration quantity for each target product. The present invention can realize store layout site selection, product selection, and product quantity configuration. It uses spatial clustering to evaluate the spatial distribution of store site selection factors, calculates the order quantity of the fishing net area, and provides a more reasonable network layout.

The invention discloses a cross-modal sensing system and method for three-dimensional positioning and tracking of a high-speed dynamic object. The system comprises an event processor, a depth processor and a factor graph optimization module. The event processor analyzes the asynchronous event flow of the event camera in real time, and generates a target two-dimensional plane position through dynamic threshold filtering and grid clustering; a depth processor processes the data stream of the depth camera, generates an ROI through coordinate projection after receiving the two-dimensional position, and analyzes and extracts target depth information by using a depth histogram; the factor graph optimization module constructs a unified optimization framework containing an event projection residual factor, a depth measurement residual factor and a motion constraint factor, and the six-degree-of-freedom pose of the target is jointly optimized through maximum posteriori estimation. The event and depth processor adopts a parallel processing mechanism, realizes heterogeneous data tight coupling through a factor graph module, and can output a sub-millisecond-level low-delay three-dimensional positioning result in a high-speed dynamic scene.

The present invention relates to an intelligent blind restoration method for low-SNR blurred star maps under high-dynamic conditions, belonging to the field of star map restoration technology. First, from the image perspective, Gaussian filtering is used to denoise the star map, thereby improving the signal-to-noise ratio of the star map; then, a grid clustering method is used to remove noise points, so that low-SNR and broken stars are effectively extracted, and the star point detection rate for large-field star sensors exceeds 90%; finally, a deep learning algorithm is used to achieve end-to-end restoration of local blurred star maps. This invention effectively solves the problems of traditional algorithms that rely on inertial navigation information, have poor real-time performance, and have a single applicable situation when dealing with the problem of restoring high-dynamic and low-SNR blurred star maps, and greatly improves the accuracy, robustness, and reliability of autonomous starlight navigation for near-space aircraft and high-angular-velocity spacecraft.

The present invention discloses a method and system for converting a building real-scene model into a monomer model, the method comprising: selecting a single building to be processed from a building real-scene three-dimensional model, obtaining a Mesh model of the single building and performing filtering processing; performing mesh clustering division on the filtered Mesh model, and extracting and simplifying the wireframe structure according to the mesh clustering result to obtain a simplified wireframe model; performing fitting projection according to the simplified wireframe model to obtain a plane polygon corresponding to each cluster cluster, and reconstructing a polygonal Mesh model; performing self-intersection polygon repair on the polygonal Mesh model through line segment intersection detection to obtain a simplified polygonal Mesh model. The present invention can automatically realize the construction of a building from a Mesh model to a monomer model, reduce the geometric complexity of the monomer building while maintaining the structure of the urban building grid model, and improve the accuracy of simplification.

The invention provides a spherical push-broom area overlapped pixel scatter data packet sequence grid clustering method and system, and the method comprises the steps: a pixel scatter data packet caching and merging step: caching a current pixel scatter data packet Pi, i being a pixel scatter data packet sequence number, and merging the current pixel scatter data packet Pi with a pixel packet # imgabs0 # missed for (i-1) times to form Pig, the # imgabs1 # being an empty packet; a spherical equal-view-angle grid clustering step: establishing grid angular points on a spherical area, establishing a projection plane through the pixel scatter point data packet, dynamically generating an auxiliary plane on the projection plane, projecting the grid angular points and the pixel scatter points to the auxiliary plane, and then performing clustering; and a missed pixel packet caching step: forming pixels in the target grids which are not clustered into a Pimiss packet for caching. According to the method, targeted measures are taken for the problem of overlapping of pixel scatter coverage areas in the front and back data packets, and a large amount of required memory resources are saved.

The invention discloses a shaft part machining process automatic generation method and system based on a large model and view recognition. The system comprises a DXF file analysis module, a view recognition module, a feature extraction engine, a process knowledge base management module, a large model interaction and process generation module and a document automatic output module. Wherein the DXF file analysis module is used for reading and analyzing a drawing in a DXF format; the view recognition module is used for recognizing a plurality of views in a drawing and classifying primitive entities by adopting a hybrid algorithm based on grid clustering and rule judgment; the feature extraction engine is used for extracting geometric features and text features of the shaft parts; the process knowledge base management module is used for storing historical process data and providing case retrieval based on cosine similarity; the large model interaction and process generation module is used for driving a large language model to generate and optimize a processing process by constructing a structured cue word; and the document automatic output module is used for generating the standardized DOCX format process document. According to the method, the engineering drawing can be automatically analyzed, the features can be intelligently recognized, the high-quality machining process can be generated, automation and intelligence of process design are achieved, and the process design efficiency and quality are improved.

The present application belongs to the field of intelligent transportation technology, and in particular to a high-precision spatio-temporal trajectory recovery method based on cell phone signaling data, using a density clustering algorithm to pre-process original cell phone signaling data, using a grid clustering algorithm that fuses spatio-temporal features, introducing POI data from Baidu maps, and carrying out a clustering analysis from the spatio-temporal two dimensions, extracting a plurality of staying points of the signaling user, and then reconstructing a signaling user's travel trajectory based on the staying points, the method provided by the present application enables the user's reconstructed trajectory to have a higher degree of precision, and is more closely fitted to the user's real trajectory.

This disclosure presents a method and apparatus for processing non-uniform image data based on a visual language model. One specific implementation of the method includes: performing text box localization processing on the image to be processed to obtain a set of text box coordinate information; obtaining a density matrix; performing grid clustering processing on the density matrix to obtain an initial set of region bounding boxes; performing density adaptive expansion processing on the initial set of region bounding boxes to obtain a set of target region bounding boxes; performing multi-scale regional extraction processing on the image to be processed to obtain a global thumbnail and a set of local sub-images; performing fusion inference processing on the global thumbnail and the set of local sub-images to obtain a target content information set; performing differential compression processing on the image to be processed to obtain archived image data; and saving the archived image data to a storage medium. This implementation can reduce the lexical consumption of visual language model inference, reduce computational resource consumption, and reduce the storage space occupied by images.